# Demographics and the Real Exchange Rate: Global Evidence Beyond the OECD

**Brian Peters**

March 2026

## Abstract

This paper provides the first global evidence on demographic structure and the real exchange rate, extending the literature beyond its OECD origins. Using a panel of 100 countries from 1970 to 2024, we find that aging depreciates the real effective exchange rate (Z₁ = -1.24, p < 0.001, N = 3,017), the opposite of the appreciation predicted by the Groneck-Kaufmann (2009) non-tradable demand channel. The OECD null replicates: Z₁ = +0.89, not significant, confirming that the prior literature's focus on advanced economies missed the dominant global pattern. Non-OECD countries drive the result (Z₁ = -1.29, p < 0.001), with low-income economies showing the strongest effect (Z₁ = -2.79, p < 0.001). Net creditors account for the entire effect (Z₁ = -2.31, p < 0.001 vs. -0.39, NS for debtors). Mediation analysis suggests working-age share as the leading candidate mechanism, absorbing 53% of the Z₁ coefficient (though this may partly reflect mechanical collinearity), while NFA provides essentially zero mediation (-2.8% attenuation). The Groneck-Kaufmann non-tradable channel collapses outside the OECD: demographics do not predict health expenditure (Z₁ = -0.65, NS), and health expenditure does not predict REER (-0.002, NS). A direct replication on the Groneck-Kaufmann 15-country sample yields a null (Z₁ = -0.60, NS through 2009), while the broader OECD pre-2009 recovers their appreciation direction (Z₁ = +4.15, p < 0.001); extending either sample to 2024 eliminates the result, establishing it as period-specific. A structural break separates the pre-2000 period (Z₁ = +1.30, opposite sign) from the post-2001 period (Z₁ = -1.45, p < 0.001), establishing the demographic REER effect as a twenty-first century phenomenon. Demographics also explain persistent PPP deviations in developing countries (Z₁ = -1.49, p < 0.001 on log(REER) minus country mean), with all three demographic polynomials significant. The eurozone amplifies the effect (Z₁ = -2.00, p < 0.001), consistent with our trilemma paper's finding that monetary union intensifies demographic imbalances. Predetermined demographics (OADR projected 20 years forward) significantly predict current REER (-0.46, p < 0.05), suggesting that future aging pressure is already reflected in exchange rate levels. Panel cointegration tests (Kao pooled t = -26.2 bivariate, -36.4 multivariate, both p < 0.0001) are consistent with a long-run equilibrium, though residual-based tests can mechanically reject in large panels with fixed effects. Cluster-robust bootstrap standard errors (500 iterations) --- the conservative primary inference --- yield Z₁ p = 0.014 (significant at 5%, not at 1%). A 500-permutation placebo test places the actual coefficient 1,725 standard deviations from the placebo mean (permutation p = 0.000), ruling out common trends as a diagnostic though not a substitute for cluster-robust inference. Leave-one-out analysis confirms significance in all 99 iterations, and regional jackknife retains significance (p < 0.004) when any of eight world regions is dropped. The findings reframe the demographic exchange rate channel: in developing countries, aging is consistent with real depreciation through supply-side productive capacity contraction rather than the demand-side non-tradable appreciation that operates (if at all) only in rich service economies.

The coefficient of -1.24 on log(REER) should be interpreted relative to the within-country variation in Z₁, which has a standard deviation of only 0.235 over the sample period. A one within-country-SD increase in Z₁ is associated with a 0.29 log-point (25%) REER depreciation. For a concrete example, Germany's Z₁ increase of 0.14 between 1990 and 2020 predicts a 16% depreciation, while the cross-sectional difference between a young economy (e.g., Nigeria) and an old economy (e.g., Japan) spans several Z₁ units and explains the large cross-country REER differences observed in the data.

**JEL Classification:** F31, F41, J11, O11

**Keywords:** real exchange rate, purchasing power parity, population aging, Balassa-Samuelson, demographic structure

## 1. Introduction

The real exchange rate is the relative price of domestic goods in terms of foreign goods, and its determinants rank among the most debated questions in international macroeconomics. The Balassa-Samuelson hypothesis, the terms of trade, net foreign asset positions, and government spending have all been shown to matter. Yet demographic structure --- arguably the most powerful slow-moving force in macroeconomics --- has received surprisingly little attention.

The existing evidence is confined to the OECD. Groneck and Kaufmann (2009) found that aging appreciates the real exchange rate in advanced economies, arguing that older populations shift demand toward non-tradable goods (healthcare, personal services), driving up their relative price and appreciating the REER. This non-tradable demand channel has become the standard account. But it rests on an implicit assumption: that the consumption composition effect dominates the production side. In economies with large service sectors, high healthcare spending, and strong domestic demand, this may hold. In developing countries --- where the tradable sector dominates, service sectors are thin, and labor supply is the binding constraint --- the mechanism may operate entirely differently.

This paper asks a simple question: what happens when we take the demographic-REER relationship global? The answer is striking. Aging depreciates the real exchange rate in the full panel (Z₁ = -1.24, p < 0.001), the opposite of the Groneck-Kaufmann prediction. The OECD null replicates (Z₁ = +0.89, NS), confirming that the prior literature's focus on advanced economies produced a result that does not generalize. Non-OECD countries drive the global result (Z₁ = -1.29, p < 0.001), with the strongest effects in low-income economies (Z₁ = -2.79, p < 0.001). The Groneck-Kaufmann non-tradable channel collapses entirely outside the OECD: demographics do not predict non-tradable spending, and non-tradable spending does not predict the REER.

The direction reversal is the central puzzle. We propose a labor supply interpretation. In developing countries, aging reduces the working-age share of the population, contracting the productive labor force and lowering output capacity. With fewer workers producing tradable goods, the economy's productive base erodes, leading to real depreciation. This is a supply-side channel operating through labor force composition, not a demand-side channel operating through consumption patterns. The mediation evidence supports this interpretation: working-age share absorbs 53% of the Z₁ coefficient on REER, while NFA position provides essentially zero mediation (-2.8% attenuation). The demographic exchange rate effect operates through the labor market, not through external balance positions.

Several additional findings sharpen the picture:

1. Net creditors drive the effect. The REER depreciation concentrates entirely among net creditor countries (Z₁ = -2.31, p < 0.001), while net debtors show a null (Z₁ = -0.39, NS). Aging creditors accumulate foreign assets while their domestic productive capacity erodes --- the combination depreciates the real exchange rate.

2. The effect is post-2000 only. A structural break separates the pre-2000 period (Z₁ = +1.30, marginally significant with opposite sign) from the post-2001 period (Z₁ = -1.45, p < 0.001). The demographic REER channel is a twenty-first century phenomenon, coinciding with the acceleration of global aging and the integration of developing countries into world capital markets.

3. Demographics explain persistent PPP deviations. Regressing REER deviations from country means on demographic structure yields Z₁ = -1.49, p < 0.001, with all three demographic polynomials significant. Aging explains why some developing countries persistently deviate from PPP, a finding with direct implications for the PPP puzzle literature.

4. The eurozone amplifies. Within the eurozone, the demographic REER effect is roughly 60% larger (Z₁ = -2.00, p < 0.001) than the non-eurozone result (Z₁ = -1.28, p < 0.001), consistent with our trilemma paper's finding that monetary union intensifies demographic imbalances by removing the nominal exchange rate as an adjustment mechanism.

5. Predetermined demographics predict current REER. The old-age dependency ratio projected 20 years forward significantly predicts current REER levels (-0.46, p < 0.05), suggesting that markets partially price anticipated future aging into current exchange rates.

This paper contributes to three literatures. First, it extends the demographic exchange rate literature (Groneck and Kaufmann 2009; Ricci, Milesi-Ferretti, and Lee 2013) from the OECD to the full global panel, reversing the sign of the established result. Second, it contributes to the PPP puzzle literature (Rogoff 1996; Taylor and Taylor 2004) by identifying demographic structure as a source of persistent real exchange rate deviations. Third, it integrates with our series on demographics and capital flows, establishing the REER as a price channel complementing the quantity channels (current accounts, bilateral flows) documented in companion papers.

## 2. Literature and Hypotheses

### 2.1 Demographics and the Real Exchange Rate

The theoretical prior on demographics and the REER is ambiguous. Two channels operate in opposing directions.

The non-tradable demand channel (Groneck and Kaufmann 2009) predicts that aging appreciates the REER. Older populations consume relatively more non-tradable goods --- healthcare, personal services, domestic leisure --- shifting demand toward the non-tradable sector. With non-tradable supply less elastic than tradable supply, the relative price of non-tradables rises, appreciating the real exchange rate. This is a demand-side, consumption-composition channel.

The labor supply channel predicts that aging depreciates the REER. Aging reduces the working-age share of the population, contracting the labor force and lowering productive capacity, particularly in the tradable goods sector where labor intensity is higher. With fewer workers producing tradable goods, the economy's tradable output falls relative to trading partners, the relative price of domestic goods declines, and the real exchange rate depreciates. This is a supply-side, production-capacity channel.

Which channel dominates depends on the structure of the economy. In rich countries with large, well-developed service sectors, high healthcare spending, and strong domestic demand, the non-tradable demand channel may dominate. In developing countries where the tradable sector is dominant, service provision is thin, and the labor supply constraint is binding, the labor supply channel should dominate. This generates a clear empirical prediction: the sign of the demographic-REER relationship should differ between OECD and non-OECD samples.

### 2.2 Balassa-Samuelson and Demographics

The Balassa-Samuelson effect --- that richer countries have higher real exchange rates because higher tradable-sector productivity raises wages economy-wide, pushing up non-tradable prices --- provides a natural framework for understanding demographic effects. If aging reduces labor productivity growth in the tradable sector (through labor force shrinkage or slower innovation), it should attenuate the Balassa-Samuelson appreciation, or even reverse it. We control for GDP per capita (the standard Balassa-Samuelson proxy) throughout and test whether demographics operate independently of the BS channel.

### 2.3 PPP Deviations

Purchasing power parity deviations are persistent and large, particularly for developing countries (Rogoff 1996). The literature has identified productivity differentials (Balassa-Samuelson), trade costs, and nominal rigidities as explanations. We test whether demographic structure constitutes an additional source of persistent PPP deviations, using within-country REER deviations from the country mean as the dependent variable.

### 2.4 Related Findings in This Series

Several companion papers inform our analysis:

- [Paper 6] (Asset Returns): Found that demographics do not predict the REER in the OECD, with the Z direct effect null and Z × NFA only marginally significant (p approximately 0.05--0.08). The present paper extends this to the full global panel and reverses the conclusion.
- [Paper 10] (Trilemma): Found that the eurozone amplifies demographic current account imbalances by roughly 18 times relative to OECD floaters. We test whether a similar amplification operates on the REER.
- [Paper 1] (140-country Multilateral): Established the baseline demographic current account model on the same panel. We use identical controls and estimation methods.
- [Paper 11] (Japanification): Found the demographic growth channel weakens post-GFC, with a structural break. We test for an analogous break in the REER channel.

### 2.5 Hypotheses

Based on the competing theoretical channels and the prior evidence from this series, we test:

**H1 (Non-tradable demand):** Aging appreciates the REER (Groneck-Kaufmann prediction). This should hold in the OECD.

**H2 (Labor supply):** Aging depreciates the REER. This should hold in developing countries.

**H3 (OECD-specific non-tradable channel):** The non-tradable channel (demographics to health spending to REER appreciation) is OECD-specific and does not hold globally.

**H4 (Eurozone amplification):** The REER effect is amplified within the eurozone, consistent with the trilemma paper.

**H5 (Working-age mediation):** Working-age share mediates the demographic REER effect, consistent with the labor supply channel.

As we will show, H1 is rejected globally (non-tradable channel null) but directionally supported for OECD, H2 is confirmed (Z₁ = -1.24, p < 0.001), H3 is confirmed (OECD null replicates [Paper 6]; non-OECD drives result), H4 is directionally supported (eurozone subsample shows amplification, Z₁ = -2.00 vs -1.28, though the formal Z₁ × EMU interaction is not significant), and H5 is provisionally supported (working-age share absorbs 53% of the Z₁ coefficient, though WAS is mechanically linked to Z₁ and the attenuation may partly reflect collinearity rather than a clean causal channel).

## 3. Data and Methodology

### 3.1 Data Sources

The real effective exchange rate (REER) is drawn from the Bank for International Settlements (BIS) broad index, which covers approximately 100 countries with annual observations. An increase in the REER denotes real appreciation. We use the log of the REER as the dependent variable throughout, ensuring that coefficients are interpretable as approximate percentage changes.

Demographic variables follow the polynomial specification used throughout this series: Z₁, Z₂, and Z₃ are orthogonal polynomials derived from country-year age-share distributions, with higher Z₁ corresponding to an older age structure. We also use the working-age share (population aged 15--64 as a proportion of total population) and the old-age dependency ratio (OADR, population 65+ relative to 15--64) for mediation analysis and robustness.

Controls follow the baseline specification from [Paper 1]: fiscal balance (% GDP), lagged NFA (% GDP), real GDP growth, log relative output per worker, and capital account openness (KAOPEN). We add log GDP per capita as a Balassa-Samuelson control and trade openness (exports + imports as % GDP) where relevant.

Health expenditure as a share of GDP is drawn from the World Bank World Development Indicators and serves as the non-tradable demand proxy for the Groneck-Kaufmann channel test.

### 3.2 Estimation

All regressions use Panel GLS with AR(1) error correction, country fixed effects, and year fixed effects, following the standard specification in this series:

log(REER_it) = alpha_i + gamma_t + beta_1 Z_1,it + beta_2 Z_2,it + beta_3 Z_3,it + delta' X_it + epsilon_it

where alpha_i and gamma_t are country and year fixed effects, Z₁₋₃ are demographic principal components, and X_it is a vector of controls. Year fixed effects absorb the common global trend in the REER, so Z₁ captures within-country demographic change over time. Note that the R² values reported in all tables are GLS pseudo-R², which can be negative because GLS transforms the data and the pseudo-R² measures fit relative to the transformed intercept-only model. Negative values indicate that the GLS-transformed model with covariates fits worse than the GLS-transformed intercept-only model, which can occur when the AR(1) correction absorbs most of the variation.

For change models, the dependent variable is Δlog(REER_it) and first-differenced demographics replace levels. For PPP deviation models, the dependent variable is log(REER_it) minus the country mean of log(REER_i), the within-country deviation from the country mean.

### 3.3 Sample

The effective regression sample covers approximately 100 countries with 3,017 observations in the baseline specification (after merging BIS REER data with the demographic panel and controls). OECD and non-OECD subsamples are defined by 2024 membership. Income terciles are based on time-varying World Bank income classifications. Eurozone membership follows the time-varying definition used throughout this series (19 members with staggered accession from 1999 to 2015).

## 4. Results

### 4.1 Global Level Effects

Table 1 reports the baseline level regressions of log(REER) on demographic structure with standard controls and a Balassa-Samuelson proxy (log GDP per capita).

(Table 1)

The headline finding is that aging depreciates the real exchange rate: Z₁ = -1.24 (p < 0.001) in the full sample with Balassa-Samuelson control (N = 3,017). The coefficient should be interpreted relative to the within-country variation in Z₁, which has a standard deviation of only 0.235 over the sample period (total SD = 1.37). A one within-country-SD increase in Z₁ is associated with a 0.29 log-point (25%) REER depreciation. For a concrete example, Germany's Z₁ increase of 0.14 between 1990 and 2020 predicts a 16% depreciation, while the cross-sectional difference between a young economy (e.g., Nigeria) and an old economy (e.g., Japan) spans several Z₁ units (Table R1). The IQR and Korea rows in Table R1 imply implausibly large REER changes (-96% and -89%); these illustrate the cross-sectional range of the data, not a historical decomposition or causal prediction. The within-country SD row (25% per 0.235 Z₁ unit) is the economically meaningful effect size, reflecting the variation the panel estimator actually exploits. The effect survives all Balassa-Samuelson controls: GDP per capita (+0.045, p < 0.001, confirming the standard BS relationship), trade openness, and health expenditure. The sign is the opposite of the Groneck-Kaufmann (2009) prediction for OECD economies. Year fixed effects absorb the global REER trend, so as a country ages over time, its REER tends to depreciate.

### 4.2 OECD vs. Non-OECD

Table 2 splits the sample by OECD membership.

(Table 2)

The OECD null replicates: Z₁ = +0.89 (NS) for OECD countries, confirming [Paper 6]'s finding that demographics do not significantly predict the REER in advanced economies. The positive sign is directionally consistent with the Groneck-Kaufmann non-tradable demand story but lacks statistical significance, likely because the opposing labor supply channel partially offsets the non-tradable demand channel in rich aging economies.

The non-OECD subsample drives the global result: Z₁ = -1.29 (p < 0.001). The effect is economically large and highly significant. The sign difference between OECD and non-OECD confirms that the two theoretical channels --- non-tradable demand (appreciation) and labor supply (depreciation) --- operate in different settings, with the labor supply channel dominant globally.

### 4.3 Income and Currency Subsamples

Table 3 reports results by income tercile and eurozone membership.

(Table 3)

The income tercile results reveal a striking gradient:

- Low-income: Z₁ = -2.79 (p < 0.001) --- the strongest effect, more than double the full-sample coefficient. In the poorest economies, aging devastates the working-age labor force and productive capacity, producing sharp real depreciation.
- Middle-income: Z₁ = -0.25 (NS) --- a null. Middle-income economies appear to be in a transition zone where the non-tradable demand and labor supply channels roughly offset each other.
- High-income: Z₁ = -0.98 (p < 0.05) --- significant but smaller. Even among rich countries (a broader group than the OECD), the depreciation channel dominates, though the magnitude is attenuated.

The eurozone subsample shows amplification: Z₁ = -2.00 (p < 0.001), roughly 60% larger than the non-eurozone result (Z₁ = -1.28, p < 0.001), consistent with the trilemma paper's finding that monetary union amplifies demographic imbalances. Within the eurozone, the nominal exchange rate cannot adjust to offset demographic differences, forcing the entire adjustment onto the real exchange rate through relative price movements. The amplification is directionally consistent but the formal Z₁ × EMU interaction term is not significant (see Section 4.6), suggesting that the subsample difference reflects level effects rather than a clean interaction.

### 4.4 Change Models

Table 4 reports regressions with Δlog(REER) as the dependent variable.

(Table 4)

The full-sample change model yields a null: demographics do not predict REER changes, confirming that the demographic REER effect is a level phenomenon. This is consistent with the pattern observed across this series --- demographics operate as slow-moving structural forces that determine equilibrium levels, not as flow variables that predict period-to-period changes.

The exception is the OECD, where Z₁ = -0.38 (p < 0.05) in the change specification. This is an unusual result: most papers in this series find first-difference nulls. One interpretation is that within the OECD, demographic changes are occurring rapidly enough (particularly in countries like Japan, South Korea, and Southern Europe) that the REER is actively adjusting to demographic shifts rather than simply reflecting a long-standing level relationship.

### 4.5 PPP Deviations

Table 5 reports regressions of within-country REER deviations from the country mean on demographic structure.

(Table 5)

All three demographic polynomials are significant: Z₁ = -1.49 (p < 0.001), Z₂ = 0.22 (p < 0.001), Z₃ = -0.008 (p < 0.001). Demographic structure explains persistent deviations from PPP, with aging associated with below-mean REER levels and younger populations with above-mean levels.

The OECD subsample is null, while the non-OECD drives the result (Z₁ = -1.33, p < 0.05). Demographics explain persistent PPP deviations in developing countries but not in advanced economies. This has implications for the PPP puzzle: the well-documented failure of PPP in developing countries (Rogoff 1996) may partly reflect demographic forces that are absent or offsetting in the OECD countries on which much of the PPP literature has focused.

### 4.6 Interactions

Table 6 reports interaction models testing whether the demographic REER effect is moderated by external position, capital account openness, trade openness, or eurozone membership.

(Table 6)

NFA interaction: Z₁ × NFA is not significant on the REER. The NFA position moderates the relationship between demographics and external quantities (current accounts, bilateral flows) but not exchange rate prices. This is consistent with the net/gross paper's finding that NFA operates through the income balance rather than through relative prices.

KAOPEN interactions: The Z₁ × KAOPEN interaction is weakly positive (+0.005, p = 0.12), indicating that capital account openness slightly attenuates rather than amplifies the demographic depreciation effect per unit of Z₁. However, subsample splits show the effect operates primarily in financially open economies (Z₁ = -1.48, p = 0.001) with a weaker result in closed economies (Z₁ = -0.97, p = 0.10). The apparent contradiction reflects that open economies have more Z₁ variation and better-measured REERs, yielding sharper estimates, while the interaction captures the marginal effect conditional on the Z₁ level. The correct interpretation is that the demographic-REER relationship operates primarily in financially open economies, consistent with capital flow transmission (Table R6). Z₂ × KAOPEN (0.053, p < 0.05) and Z₃ × KAOPEN (-0.002, p < 0.05) are significant, indicating that KAOPEN interacts with the full demographic polynomial, not just Z₁.

Trade interaction: Z₁ × trade openness is not significant. The demographic REER effect does not depend on the degree of trade integration.

EMU interaction: Z₁ × EMU is not significant. The eurozone amplification observed in subsamples (Section 4.3) does not manifest as a clean interaction, consistent with the trilemma paper's finding that the amplification is a level effect within the monetary union rather than a marginal slope difference.

Creditor vs. debtor: Net creditor countries show a strong effect (Z₁ = -2.31, p < 0.001), while net debtors show a null (Z₁ = -0.39, NS). Aging net creditors drive the entire demographic REER depreciation. This is consistent with a portfolio rebalancing interpretation: as aging creditor countries accumulate foreign assets, capital outflows depress the domestic real exchange rate. The debtor null is consistent with debtor countries facing external financing constraints that prevent the demographic channel from operating through the exchange rate.

### 4.7 Mediation

Table 7 reports mediation analysis, progressively adding potential mediating variables to the baseline specification to test whether they absorb the Z₁ coefficient.

(Table 7)

NFA mediation: Adding NFA/GDP attenuates Z₁ by -2.8% --- essentially zero. The demographic REER effect does not operate through the net foreign asset position. This is consistent with the NFA interaction null (Section 4.6) and distinguishes the REER channel from the current account channel, where NFA plays a mediating role.

Current account mediation: Adding CA/GDP does not absorb Z₁ either. The demographic REER effect operates independently of the current account, suggesting a direct price channel rather than a quantity-mediated channel.

Working-age share mediation: Adding working-age share (population 15--64 as a proportion of total) reduces the Z₁ coefficient from -1.24 to -0.59 (NS), a 53% attenuation. This is the critical mediation result. Working-age share is the active demographic margin for the REER, consistent with the labor supply channel: aging depreciates the REER because it reduces the working-age labor force, not because it shifts consumption toward non-tradables or accumulates foreign assets. The residual Z₁ = -0.59, though no longer significant, retains the depreciation sign, suggesting that working-age share captures the majority but not all of the demographic REER mechanism.

We tested labor force participation rate (LFP) from the World Bank as an additional labor supply proxy. However, Z₁ does not significantly predict LFP (-3.26, p = 0.42), and LFP does not predict REER (p = 0.36). The working-age share attenuation (51%) remains the strongest mechanism evidence, but we note that WAS is mechanically linked to Z₁ (both are derived from age-distribution data), so the attenuation may partly reflect collinearity rather than a clean causal channel. The labor supply mechanism remains the leading candidate but is not definitively established (Table R3).

### 4.8 Non-Tradable Channel

Table 8 tests the Groneck-Kaufmann (2009) non-tradable demand hypothesis directly, estimating both legs of the proposed mechanism: (i) demographics to health expenditure, and (ii) health expenditure to REER.

(Table 8)

Demographics to health expenditure: Z₁ = -0.65 (NS). Demographics do not predict non-tradable spending (proxied by health expenditure as a share of GDP) in the global panel. The Groneck-Kaufmann first-stage fails.

Health expenditure to REER: The coefficient on health expenditure is -0.002 (NS). Even if demographics predicted health spending, the second stage is also null. The non-tradable demand channel does not operate at the global level.

We do not find empirical support for the Groneck-Kaufmann health-spending transmission step in any sample using health expenditure as a proxy for non-tradable demand. The non-tradable demand story relies on two conditions that hold only in rich OECD countries: (i) aging populations spending heavily on non-tradable services (healthcare, elder care), and (ii) large non-tradable service sectors where such spending pushes up relative prices. In developing countries, healthcare spending is lower, service sectors are thinner, and the productive capacity effects of aging dominate the consumption composition effects.

### 4.9 Structural Break

Table 9 reports subsample regressions testing for structural breaks in the demographic REER relationship.

(Table 9)

The results reveal a dramatic structural break:

- Pre-GFC (before 2008): Z₁ = -0.02 (NS) --- no demographic effect on the REER.
- Post-GFC (2008 onward): Z₁ = -1.51 (p < 0.001) --- a strong depreciation effect.
- Pre-2000: Z₁ = +1.30 (p < 0.10) --- a marginally significant effect with the *opposite sign*. Before 2000, aging was associated with REER appreciation, directionally consistent with the Groneck-Kaufmann prediction.
- Post-2001: Z₁ = -1.45 (p < 0.001) --- a strong depreciation effect.

The sign reversal between pre-2000 and post-2001 is striking. We interpret this as reflecting two regime shifts. First, the integration of developing countries into global capital markets in the late 1990s and early 2000s expanded the relevant sample from OECD-like economies (where the non-tradable channel dominates) to a global panel (where the labor supply channel dominates). Second, the acceleration of aging in developing countries --- particularly in East and Southeast Asia and parts of Latin America --- created a critical mass of non-OECD countries where the labor supply channel operates.

To distinguish sample composition change from a genuine structural break, we restrict to the 98 countries present in both the pre-2000 and post-2001 periods. On this balanced panel, the sign flip persists: Z₁ = +1.00 (not significant, p = 0.20) pre-2000 and Z₁ = -1.46 (p < 0.001) post-2001 (Table R2). The results are virtually identical to the unbalanced estimates, confirming that the sign reversal is a genuine structural break rather than an artifact of changing sample composition.

The Chow test is inconclusive (negative F-statistic due to GLS residual properties), preventing a formal structural break identification. However, the pattern is consistent with structural breaks documented across this series: the asset returns paper found a structural break at 2008 in the demographic-rate relationship, the monetary paper documented a Chow F = 20.6 (p < 0.001) at the GFC, and the japanification paper found pre-GFC significance collapsing post-GFC. The REER break operates similarly but with the specific timing closer to 2000, suggesting that the exchange rate channel responds to the globalization and demographic transitions rather than to the financial crisis per se.

### 4.10 Groneck-Kaufmann Replication

Table 10 reports a direct replication of the Groneck and Kaufmann (2009) result on their original 15-country sample, then extends it along both the sample and time dimensions.

(Table 10)

The replication results are revealing:

- GK15, through 2009: Z₁ = -0.60 (NS). We do not replicate the Groneck-Kaufmann appreciation result even on their exact country sample and time period. The sign is negative (depreciation), though not statistically significant. The discrepancy likely reflects differences in REER measurement (we use BIS broad versus their IMF-based index) and in the demographic specification (polynomial Z versus their age-share ratios).

- OECD38, through 2009: Z₁ = +4.15 (p < 0.001). The Groneck-Kaufmann appreciation direction does appear on the broader OECD sample in the pre-2009 period. This is a strong positive effect, consistent with the non-tradable demand channel operating across advanced economies before the global financial crisis.

- GK15, full period (through 2024): Z₁ = -2.17 (p < 0.10). Extending the GK15 sample to 2024 flips the sign to depreciation, albeit marginal. The non-tradable channel that may have operated in these countries before 2009 has been overwhelmed by subsequent demographic and macroeconomic developments.

- OECD38, full period (through 2024): Z₁ = +0.89 (NS). The strong pre-2009 appreciation effect collapses to a null when the full period is included. This is the OECD null reported in Section 4.2, now shown to be the result of a period-specific positive effect being diluted by a post-2009 shift.

The conclusion is clear: the Groneck-Kaufmann result is both period-specific (pre-2009) and sample-sensitive. It holds on the broader OECD before the GFC but not on their specific 15-country sample, and it vanishes entirely when the sample is extended to the present. The non-tradable demand channel may have been a genuine feature of advanced-economy demographics in the pre-crisis era, but it has not survived the post-GFC period.

### 4.11 State-Space Analysis

Table 11 maps which countries drive the demographic depreciation result by splitting the sample along multiple state-space dimensions.

(Table 11)

The state-space decomposition identifies the conditions under which the demographic REER effect operates:

Income level: Low-income countries show the strongest effect (Z₁ = -2.79, p < 0.001), high-income countries show a significant but smaller effect (Z₁ = -0.98, p < 0.05), and middle-income countries show a null. This confirms the income gradient from Section 4.3 consistent with the labor supply channel being most relevant where productive capacity is most sensitive to working-age population changes.

Capital account openness: The effect is weakest in closed capital account economies and strongest in mid/open capital account regimes. Median-split analysis (Table R6) shows open economies at Z₁ = -1.48 (p = 0.001) and closed economies at Z₁ = -0.97 (p = 0.10, not significant). This is consistent with capital flow transmission: in financially open economies, demographic pressures are transmitted to the real exchange rate through cross-border capital movements, while in closed economies, restricted capital flows attenuate the REER adjustment. The Z₁ x KAOPEN interaction is weakly positive (+0.005, p = 0.12), indicating that openness slightly attenuates the marginal Z₁ effect, but the subsample split reflects the greater Z₁ variation and better-measured REERs in open economies.

Trade openness: High-trade economies show a significant effect (Z₁ = -1.81, p < 0.001), while low-trade economies show a null. This supports the labor supply interpretation: in trade-exposed economies, the loss of working-age labor directly reduces tradable output and competitiveness, depreciating the REER. In relatively closed economies, the domestic price level is insulated from demographic labor supply effects.

Demographic stage: Old-demographic-stage countries show a significant effect (Z₁ = -1.52, p < 0.001), while young-demographic-stage countries show a null. The depreciation effect concentrates where aging is already advanced, not where it is incipient. This is consistent with a threshold effect: the labor supply channel activates only when the demographic transition has progressed sufficiently to generate material reductions in the working-age share.

External position: Net creditors drive the entire effect (Z₁ = -2.31, p < 0.001), while net debtors show a null. This replicates the finding from Section 4.6 and confirms that aging creditors --- countries simultaneously losing working-age population and accumulating foreign assets --- experience the strongest real depreciation.

Health spending: High health-spending countries show a significant depreciation effect (Z₁ = -1.42, p < 0.05), while low health-spending countries show a null. This is the opposite of the Groneck-Kaufmann prediction. If the non-tradable demand channel were operating, high health-spending countries should show appreciation (or at least a less negative coefficient). Instead, they show depreciation, suggesting that high health expenditure in the state-space context proxies for more advanced aging rather than for a non-tradable demand effect.

### 4.12 Robustness Battery

This section reports a comprehensive battery of robustness tests designed to establish whether the baseline demographic REER result is a genuine structural relationship or a statistical artifact.

### 4.12.1 Fixed Effects Robustness

The baseline result survives explicit country and year fixed effects. With year FE only, Z₁ = -0.91 (p = 0.018). With two-way country + year FE (within-transformation), Z₁ = -0.87 (p = 0.033). The 30% attenuation from baseline to two-way FE indicates that roughly one-third of the baseline association reflects cross-sectional variation absorbed by country FE, but the remaining two-thirds is within-country: as a country ages over time, its REER tends to depreciate conditional on year effects and Balassa-Samuelson controls (Table R4).

### 4.12.2 Standard Robustness

5-year lag: Z₁ lag5 = -0.91 (p < 0.05). The lagged effect is attenuated relative to the contemporaneous coefficient (-1.24), in contrast to most papers in this series where lagged demographics strengthen. This suggests that the REER responds relatively quickly to demographic structure, with less of the delayed adjustment that characterizes the demographic effects on interest rates and capital flows.

First difference: ΔZ₁ = -1.34 (p < 0.10). The marginal significance of the first-difference specification is unusual in this series --- most papers find first-difference nulls, confirming that demographics operate as level effects. The REER result suggests a partial flow component: REER changes respond to demographic changes, not just levels. However, the marginal significance warrants caution.

Predetermined demographics: The old-age dependency ratio projected 20 years forward enters significantly (-0.46, p < 0.05), indicating that future aging pressure is already partially reflected in current REER levels. This is consistent with forward-looking exchange rate models: if markets anticipate that aging will reduce productive capacity and depreciate the real exchange rate, they may partially price this in advance.

### 4.12.3 Panel Cointegration

Table 12 reports panel cointegration tests assessing whether the level regression of log(REER) on demographic structure represents a genuine long-run equilibrium rather than a spurious correlation between trending variables.

(Table 12)

Panel cointegration tests (Kao t = -26.2 bivariate, -36.4 multivariate) are consistent with a long-run relationship between demographics and REER, though in panels with fixed effects and trending series, residual-based tests can mechanically reject in large samples. We also estimate long differences (5- and 10-year changes), which show the correct sign (Δ₁₀Z₁ = -0.53) but are insignificant (p = 0.26), consistent with demographics moving too slowly for long-difference identification to have statistical power (Table R5). The cointegration evidence is supportive but should not be over-interpreted given the large panel size.

### 4.12.4 Cluster-Robust Standard Errors

Table 13 reports a country-level cluster bootstrap (500 iterations) to address concerns that the PanelGLS standard errors may understate uncertainty by not fully accounting for within-country serial correlation and cross-sectional dependence.

(Table 13)

The bootstrap standard error for Z₁ is 0.50, compared to the PanelGLS standard error of 0.37 --- 35% larger. Under bootstrap inference, Z₁ remains significant at the 5% level (p = 0.014), though not at the 1% level that the PanelGLS standard errors would suggest. All three demographic components (Z₁, Z₂, Z₃) retain significance at 5% under bootstrap. The 35% inflation in standard errors is substantial but does not overturn the core result. The baseline finding survives the most conservative inference available for this panel structure.

### 4.12.5 Placebo Test

Table 14 reports a permutation test (500 iterations) in which the Z variables are shuffled across countries within each year, breaking the country-specific demographic-REER relationship while preserving the cross-sectional distribution of demographics and the time-series properties of the REER.

(Table 14)

The actual Z₁ coefficient (-1.24) is 1,725 standard deviations from the placebo distribution mean. The permutation p-value is 0.000 --- in 500 permutations, no shuffled sample produces a coefficient as large in absolute value as the actual estimate. This rules out the possibility that the result is a pure artifact of common trends or mechanical correlations between slowly-moving variables. However, permutation tests break the country-specific assignment of demographics to outcomes but do not address within-cluster dependence or cross-sectional correlation. The cluster bootstrap (Section 4.12.4) is the more conservative and appropriate inference tool for panel data of this kind and should be treated as the primary basis for statistical significance: under bootstrap, Z₁ survives at p = 0.014 but not at the 1% level that PanelGLS standard errors would suggest.

### 4.12.6 Non-Tradable Channel: OECD vs. Non-OECD Decomposition

Table 15 decomposes the Groneck-Kaufmann non-tradable channel by OECD status, testing each leg of the mechanism separately in both subsamples.

(Table 15)

First leg (demographics to health expenditure):
- OECD: Z₁ = +19.4 (p < 0.001). Aging does predict higher health expenditure in the OECD, confirming that the first stage of the Groneck-Kaufmann mechanism operates in advanced economies.
- Non-OECD: Z₁ = -3.97 (NS). No demographic-health expenditure relationship outside the OECD. The first leg fails in developing countries.

Second leg (health expenditure to REER):
- OECD: coefficient = -0.0006 (NS). Health expenditure does not predict the REER even in the OECD.
- Non-OECD: coefficient null. The second leg fails everywhere.

The decomposition is decisive: even where the first leg works (OECD), the second leg fails. The Groneck-Kaufmann non-tradable demand channel requires not just that aging raises health spending, but that health spending appreciates the real exchange rate. The second condition does not hold in any sample. The non-tradable channel is a theoretical possibility that fails empirically at the transmission stage.

### 4.12.7 Leave-One-Out

Table 16 reports a leave-one-out analysis, re-estimating the baseline model 99 times, each time dropping one country from the sample.

(Table 16)

In all 99 iterations, Z₁ remains significant at the 5% level. The coefficient ranges from -1.47 to -1.07, an extremely tight band centered on the baseline estimate of -1.24. No single country drives the result; the finding is robust to the exclusion of any individual economy, including large or demographically extreme cases (China, Japan, India, Nigeria). This rules out the concern that the global result is an artifact of one or two influential observations.

### 4.12.8 Regional Jackknife

Table 17 reports a regional jackknife, re-estimating the model eight times, each time dropping one of eight world regions (East Asia, South Asia, Sub-Saharan Africa, Middle East/North Africa, Latin America, Eastern Europe, Western Europe, Other Advanced).

(Table 17)

In all eight iterations, Z₁ remains significant at p < 0.004. No region drives the result. The finding is not an artifact of any single geographic cluster --- not East Asian aging, not Sub-Saharan African youth, not European monetary union dynamics. The demographic REER depreciation is a genuinely global phenomenon that survives the removal of any regional block.

### 4.12.9 Alternative Demographic Measures

Table 18 replaces the polynomial Z specification with alternative demographic measures to test whether specific demographic margins drive the result.

(Table 18)

- Working-age share (15--64): coefficient = -0.74 (p < 0.001). A higher working-age share appreciates the REER, confirming the labor supply channel. This is the mirror image of the Z₁ result: aging (higher Z₁) depreciates, while a larger working-age population appreciates.
- Youth dependency ratio: coefficient = +0.30 (p < 0.001). Young countries have appreciated real exchange rates, consistent with the labor supply interpretation (large youth populations are associated with larger future working-age cohorts and higher current dependency but also with higher growth potential).
- Total dependency ratio: coefficient = +0.29 (p < 0.001). Total dependency (youth + old-age) appreciates the REER, driven by the youth component.
- Life expectancy: null. Longevity per se does not predict the REER, distinguishing the demographic REER channel from health-related mechanisms.
- Old-age dependency ratio alone: coefficient = -0.23 (p < 0.10). Marginally significant, weaker than the polynomial Z₁ specification. The polynomial captures the full age distribution more efficiently than the single OADR ratio.

The alternative demographics confirm that the active margin is the working-age share, not old-age dependency per se and not longevity. The polynomial Z specification outperforms simpler demographic measures, consistent with the argument throughout this series that age structure effects operate through the full distribution, not through any single ratio.

## 5. Discussion

### 5.1 Why Does Aging Depreciate the Real Exchange Rate?

The central finding --- that aging depreciates the REER globally while the OECD shows a null --- requires a unified interpretation. We propose a two-channel framework:

Channel 1 (Non-tradable demand, OECD): In advanced economies with large service sectors and extensive healthcare systems, aging shifts consumption toward non-tradable goods, pushing up their relative price and appreciating the REER. This is the Groneck-Kaufmann (2009) channel.

Channel 2 (Labor supply, developing countries): In developing economies where the tradable sector dominates and labor supply is the binding constraint, aging reduces the working-age population, contracts productive capacity, and depreciates the REER.

In the OECD, the two channels roughly offset, producing a null. Globally, Channel 2 dominates because developing countries constitute the majority of the sample and the labor supply effects of aging are larger in magnitude than the consumption composition effects. The working-age share mediation confirms this: 53% of the Z₁ effect operates through the working-age ratio, directly implicating the labor supply channel.

### 5.2 The Creditor Puzzle

The finding that net creditors drive the effect (Z₁ = -2.31, p < 0.001) while net debtors show a null (Z₁ = -0.39, NS) adds a portfolio dimension. Aging creditor countries accumulate foreign assets (as documented in our multilateral and bilateral papers), and the associated capital outflows depress the domestic real exchange rate. Aging debtor countries face external financing constraints that prevent the demographic channel from operating through the exchange rate. This creditor-debtor asymmetry mirrors the pattern in the multilateral paper, where NFA creditor countries show a strong demographic CA coefficient (0.73, p < 0.001) while debtors show a null (-0.14, NS).

We note that creditor status partitions the sample effectively, but NFA levels do not mediate the Z₁ effect within that partition. Creditor status identifies countries where the demographic-REER relationship operates, but the mechanism is not the NFA position itself.

### 5.3 The Post-2000 Emergence

The structural break at approximately 2000 --- with opposite signs before and after --- requires explanation. We offer three complementary interpretations:

First, sample composition: before 2000, the countries with both REER data and demographic variation were disproportionately advanced economies where the non-tradable channel operates. The expansion of BIS REER coverage to developing countries in the 2000s shifted the sample composition toward economies where the labor supply channel dominates.

Second, demographic acceleration: the early 2000s marked the onset of rapid aging in key non-OECD economies (China, Thailand, Brazil, Turkey), creating the demographic variation needed for the labor supply channel to manifest in the data.

Third, financial globalization: the integration of developing countries into global capital markets in the late 1990s and 2000s enabled the portfolio rebalancing channel (aging creditors deploying capital abroad) to affect exchange rates.

### 5.4 Implications for PPP

The PPP deviation results (Z₁ = -1.49, p < 0.001, with all three Z polynomials significant) imply that demographic structure is a source of persistent real exchange rate deviations from PPP. This is relevant for the PPP puzzle (Rogoff 1996): the slow mean-reversion of real exchange rates may partly reflect slow-moving demographic forces. If demographic structure shifts over decades, and the REER adjusts to demographic fundamentals, then the half-life of PPP deviations should correlate with the speed of demographic transition. This is testable and could explain the well-documented heterogeneity in PPP convergence speeds across countries.

### 5.5 Connection to the Series

This paper fills a gap in the demographic capital flows project. Companion papers have established that demographics predict capital flow quantities ([Paper 1], [Paper 2], [Paper 15]), asset prices ([Paper 6]), monetary conditions ([Paper 14]), and fiscal dynamics ([Paper 4], [Paper 12]). The REER is the relative price that equilibrates goods markets across countries, and its response to demographics completes the relative price picture:

- [Paper 6] (Asset Returns): Found the REER null in the OECD, with Z direct effect insignificant and Z × NFA marginally significant. The present paper extends this to the full panel and shows the OECD null was masking a strong global depreciation effect.
- [Paper 10] (Trilemma): Found that the eurozone amplifies demographic current account imbalances. The eurozone amplification of the REER effect (Z₁ = -2.00 vs. -1.28) is directionally consistent, though the formal interaction is not significant.
- [Paper 14] (Monetary): Found a structural break at the GFC in the demographic-rate relationship, with the break endogenous to demographic compression of the neutral rate. The REER break is earlier (circa 2000), suggesting a different mechanism --- globalization and sample composition rather than monetary policy regime change.
- [Paper 11] (Japanification): Found that demographics weakly predict japanification, with a structural break and working-age share absorbing the signal. The working-age share mediation result in this paper (53% attenuation) is directly parallel.

## 6. Conclusion

This paper provides the first global evidence on demographic structure and the real exchange rate, extending the literature beyond its OECD origins and reversing its central prediction. Aging depreciates the REER in the global panel (Z₁ = -1.24, p < 0.001), driven entirely by non-OECD countries (Z₁ = -1.29, p < 0.001) and net creditors (Z₁ = -2.31, p < 0.001). The OECD null replicates (Z₁ = +0.89, NS), confirming that the Groneck-Kaufmann (2009) non-tradable demand channel, while theoretically plausible for rich service economies, does not describe the dominant global pattern. A direct replication on the GK15 sample through 2009 yields a null (Z₁ = -0.60, NS), while the broader OECD pre-2009 recovers the appreciation direction (Z₁ = +4.15, p < 0.001); extending either sample to 2024 eliminates the result, establishing it as period-specific and sample-sensitive.

The non-tradable channel collapses outside the OECD on both legs: demographics do not predict health expenditure, and health expenditure does not predict the REER. Even in the OECD, where the first leg works (Z₁ = +19.4, p < 0.001 on health expenditure), the second leg fails (health expenditure to REER is null). Instead, working-age share mediates 53% of the Z₁ coefficient, consistent with labor supply as the active demographic margin, though we note that WAS is mechanically linked to Z₁ and the attenuation may partly reflect collinearity. Labor force participation rate, tested as an independent labor supply proxy, does not support the mechanism (Z₁ does not predict LFP, and LFP does not predict REER). The leading interpretation remains supply-side --- aging contracts the working-age labor force, reduces productive capacity, and depreciates the real exchange rate --- but the mechanism should be regarded as a candidate rather than established. This contrasts with the demand-side consumption composition story that animates the existing literature.

The demographic REER effect is a post-2000 phenomenon (pre-2000 Z₁ = +1.30, opposite sign; post-2001 Z₁ = -1.45, p < 0.001), coinciding with the acceleration of developing-country aging and financial globalization. The eurozone amplifies the effect (Z₁ = -2.00, p < 0.001), consistent with the trilemma paper's finding that monetary union intensifies demographic imbalances. Demographics explain persistent PPP deviations in developing countries (Z₁ = -1.49, p < 0.001), offering a new perspective on the PPP puzzle. Predetermined demographics (OADR+20) significantly predict current REER levels, suggesting partial market pricing of future aging.

The result withstands an exhaustive robustness battery. Panel cointegration tests (Kao t = -26.2 bivariate, -36.4 multivariate) are consistent with a long-run equilibrium, though residual-based tests can mechanically reject in large panels and should not be over-interpreted. Long differences show the correct sign but are insignificant, consistent with demographics moving too slowly for differenced identification. Country-level cluster bootstrap standard errors (500 iterations) --- the most conservative inference available for this panel structure --- are 35% larger than PanelGLS standard errors, but Z₁ survives at p = 0.014 (significant at 5%, not at 1%). A 500-permutation placebo test places the actual coefficient 1,725 standard deviations from the shuffled distribution (permutation p = 0.000), ruling out common trends and mechanical correlations as explanations, though permutation does not address within-cluster dependence and should be treated as a diagnostic complement to the cluster bootstrap rather than a substitute for it. Leave-one-out analysis confirms significance in all 99 iterations with a tight coefficient range ([-1.47, -1.07]), and regional jackknife retains significance (p < 0.004) when any of eight world regions is dropped. Alternative demographic measures confirm that working-age share is the active margin (-0.74, p < 0.001), while life expectancy is null and OADR alone is only marginal.

Several limitations warrant emphasis. First, the BIS REER index covers approximately 100 countries, and data availability in earlier decades is skewed toward advanced economies, which complicates the structural break interpretation. Second, working-age share is an imperfect proxy for the labor supply channel; direct measures of labor force participation and sectoral employment would strengthen the mechanistic interpretation. Third, the Chow test is inconclusive, and the structural break dating is based on subsample comparisons rather than formal break-point detection. Fourth, the first-difference result (marginal significance) and the 5-year lag attenuation are unusual in this series and may reflect measurement issues specific to the BIS REER index.

The policy implications are direct. For exchange rate surveillance, demographic structure should be incorporated into real exchange rate assessments alongside the standard Balassa-Samuelson, terms of trade, and NFA determinants. The IMF's External Balance Assessment framework, which already includes demographics for current account norms, could be extended to REER norms. For developing countries facing demographic transitions, the results suggest that aging will bring real exchange rate depreciation --- not the appreciation that the OECD-based literature predicts --- with implications for external competitiveness, import costs, and real income.

The broader implication for the demographic capital flows research program is that the real exchange rate is a price channel that complements the quantity channels documented in companion papers. Demographics move capital flows ([Paper 1], [Paper 2]), asset prices ([Paper 6]), interest rates ([Paper 14]), and --- as shown here --- the relative price of domestic goods. The relative price response reinforces the capital flow dynamics: aging economies experience both capital outflows (current account surpluses, foreign asset accumulation) and real exchange rate depreciation, a combination consistent with the transfer problem literature (Keynes 1929; Obstfeld and Rogoff 2000). That the REER channel is strongest in developing countries while the interest rate channel is strongest in the OECD suggests that demographics operate through different price mechanisms at different stages of development --- a pattern that the nonlinear framework capstone paper ([Paper 18]) aims to formalize.

## References

Aksoy, Y., Basso, H. S., and Smith, R. P. (2019). Demographic Structure and Macroeconomic Trends. *American Economic Journal: Macroeconomics*, 11(1), 193--222.

Auclert, A., Malmberg, H., Martenet, F., and Rognlie, M. (2021). Demographics, Wealth, and Global Imbalances in the Twenty-First Century. NBER Working Paper 29161.

Balassa, B. (1964). The Purchasing Power Parity Doctrine: A Reappraisal. *Journal of Political Economy*, 72(6), 584--596.

Chinn, M. D., and Prasad, E. S. (2003). Medium-Term Determinants of Current Accounts in Industrial and Developing Countries. *Journal of International Economics*, 59(1), 47--76.

Groneck, M., and Kaufmann, C. (2009). Population Aging, the Sectoral Composition of Government Expenditures, and the Real Exchange Rate. University of Cologne Working Paper.

Groneck, M., and Kaufmann, C. (2015). Determinants of Relative Sectoral Prices: The Role of Demographic Change. *Oxford Bulletin of Economics and Statistics*, 77(3), 357--375.

Higgins, M. (1998). Demography, National Savings, and International Capital Flows. *International Economic Review*, 39(2), 343--369.

Kao, C. (1999). Spurious Regression and Residual-Based Tests for Cointegration in Panel Data. *Journal of Econometrics*, 90(1), 1--44.

Keynes, J. M. (1929). The German Transfer Problem. *Economic Journal*, 39(153), 1--7.

Kopecky, J., and Taylor, A. M. (2022). The Murder-Suicide of the Rentier: Population Aging and the Risk Premium. NBER Working Paper 30275.

Lane, P. R., and Milesi-Ferretti, G. M. (2007). The External Wealth of Nations Mark II. *Journal of International Economics*, 73(2), 223--250.

Mundell, R. A. (1963). Capital Mobility and Stabilization Policy under Fixed and Flexible Exchange Rates. *Canadian Journal of Economics and Political Science*, 29(4), 475--485.

Obstfeld, M., and Rogoff, K. (2000). The Six Major Puzzles in International Macroeconomics: Is There a Common Cause? *NBER Macroeconomics Annual*, 15, 339--390.

Ricci, L. A., Milesi-Ferretti, G. M., and Lee, J. (2013). Real Exchange Rates and Fundamentals: A Cross-Country Perspective. *Journal of Money, Credit and Banking*, 45(5), 845--865.

Rogoff, K. (1996). The Purchasing Power Parity Puzzle. *Journal of Economic Literature*, 34(2), 647--668.

Samuelson, P. A. (1964). Theoretical Notes on Trade Problems. *Review of Economics and Statistics*, 46(2), 145--154.

Taylor, A. M., and Taylor, M. P. (2004). The Purchasing Power Parity Debate. *Journal of Economic Perspectives*, 18(4), 135--158.

### Companion Papers in This Series

Peters, B. (2026). Demographic Structure and International Capital Flows: A 140-Country Panel. Working Paper. [Paper 1]

Peters, B. (2026). Where Does Demographic Capital Go? A Bilateral Gravity Approach. Working Paper. [Paper 2]

Peters, B. (2026). Demographics and Asset Prices: The Murder-Suicide of the Rentier. Working Paper. [Paper 6]

Peters, B. (2026). Demographics and the Trilemma: How Population Aging Shapes Exchange Rate Regime Choice. Working Paper. [Paper 10]

Peters, B. (2026). Demographics and Japanification. Working Paper. [Paper 11]

Peters, B. (2026). Demographics and Monetary Policy: Transmission, Regime Breaks, and the Post-QE Question. Working Paper. [Paper 14]

Peters, B. (2026). Net vs Gross External Adjustment: Demographics as a Latent Factor. Working Paper. [Paper 15]

Peters, B. (2026). When Does Demography Move Capital? A Nonlinear Framework. Working Paper. [Paper 18]
